Differentiating densities on smooth manifolds

作者:

Highlights:

• Monte Carlo integration of derivatives of oscillatory functions is very expensive.

• For Lebesgue integrals, integration by parts gives rise to a density gradient g.

• g is defined as a directional derivative of density implied by a coordinate chart.

• Computation of g requires both the Jacobian and Hessian of the chart.

• First- and second-order tangent equations must be solved to find g on trajectories.

摘要

•Monte Carlo integration of derivatives of oscillatory functions is very expensive.•For Lebesgue integrals, integration by parts gives rise to a density gradient g.•g is defined as a directional derivative of density implied by a coordinate chart.•Computation of g requires both the Jacobian and Hessian of the chart.•First- and second-order tangent equations must be solved to find g on trajectories.

论文关键词:Density gradient function,Differentiable manifold,Lebesgue integral,Monte Carlo integration,Linear response

论文评审过程:Received 13 March 2021, Revised 14 May 2021, Accepted 6 June 2021, Available online 20 June 2021, Version of Record 20 June 2021.

论文官网地址:https://doi.org/10.1016/j.amc.2021.126444